Lecture: Technology-Enabled Eye Screening Programs: Advances, Lessons, and Limits in Community Eye Health

Thank you everyone for joining us. Certainly, I want to welcome everyone to our live webinar. We have a lot to cover today with some really interesting people and perspectives about how technology is going to be used. Anusha, I want you to come off mute and thank you for being a part of this run of show. >> Thank you, Hunter. Welcome to today’s Cybersight webinar. I’m an Anusha Purushotham. An epidemiologist and head of health systems at Remidio. We have an exciting panel today. With pleasure, I introduce our panelists for today. Dr. Nathan Congdon, ophthalmologist at the University of Belfast. Dr. Ciku Mathenge, President of African Ophthalmic Counsel. Dr. Charles Cleland from the London School of Hygiene and Tropical Medicine. And Dr. Kwesi, ophthalmologist at the University Kana Medical School. Moderating is Dr. Hunter. And before we begin the panel discussion for today, let me take a few minutes to set the stage. Around 1.1 billion people live with vision loss and roughly 90 percent of them are completely preventable. Most that live with vision loss live in low-income countries. It’s critical that screening occurs on time. This is precisely where technology can change what is achievable. AI, portable imaging and — ophthalmology have promised to put eye screening within everyone’s reach. Today we will hear from panelists on their lived experiences of using these technologies on the ground. We look forward to understanding how they chose their tools, what they learned and how technology moved the needle on program impact. I now hand the platform over to Hunter. >> I’m super excited. If you had told me 25 years ago when I was graduating medical school that we’d be talking about artificial intelligence and gene therapy, I would have told you you’re dreaming. We have people on the panel that made those dreams come true. Dr. Kwesi, I want to start with you. Before we get into all the lessons learned, tell me how you’re using AI in Rwanda. What does it look like from the nursing perspective and the patient perspective. >> Thank you Cybersight for including me on this panel. Hunter, our AI program is Rwanda is driven by a severe shortage of ophthalmologists. We don’t have enough specialists to look at all of the retinas that belong to at-risk people. And we prioritized AI as our go-to just because of that mathematical reality. So before deployment, we had a coalition of experts from the ministry of health, ministry of IT, advocacy groups, patient, stakeholders from data and legal, and they had to collaborate first so we established a supportive ecosystem. After that, we also did some research. We tested the AI to refine its accuracy specifically for African retinas and today our program operates across 4 clinics. Some in referral center, some in the private sector and the patient clinics and they are run by nurses. We wanted an efficient model so we trained them in one day: We have to choose cameras that are really easy to use and trained them how to take images and upload to Cybersight AI. For the patients to ensure comprehension and bridge the literacy gap, we put it into a customized local report. Let me tell you about a typical day in clinic. I’m talking from a nurse’s perspective. Patients arrive and go through registration and join the cue for the regular diabetes clinic. The AI nurse, let me call her that for clarity, will use this wait time to introduce the AI diabetic screening program. We actually have pull up banners in the waiting areas explaining the program. She will then invite anybody who hasn’t been screened in the past 6 months to step into the room next door, the camera is next door to the doctor’s office for screening. So at the same time for patients who are previously screened and referred to an eye specialist, she will take a moment to just follow up on what happened and remind them that they need to update their diabetes doctor and what happened. The last, we’ll then screen the patients and she will hand them, the two reports, the AI and the customized one. They take these directly to the diabetes doctor. Now, we have some clinics that have adapted this work flow depending on their setup where the nurse sometimes will only screen patients after they leave the diabetes doctor office. But we find that we lose some screenable patients when they choose that model as they leave the doctor with so many competing instructions. So that’s how I would summarize a typical screening day. >> I love that. I think in the trial and Andy is going to send this to the group, you also looked at patient and provider satisfaction. I love the fact that it actually improved patient and family satisfaction of the visit and their compliance with referral. So those diagnosed in remote areas had a higher chance or going for their treatment if they received the AI. And Nathan, I’m going to jump to you now, Dr. Congdon, and say you’ve been jumping to another disease looking at glaucoma. Can you talk about the workflow, the lessons learned and what you did in that glaucoma space. Because is that may be one of the more difficult posterior diseases for AI to diagnose. >> In any screening program, the main challenge is to bring the capacity to treat and patient together in one room. That is challenging for glaucoma because no one recognizes what glaucoma is, is in itself a major challenge. We had some work we did in Vietnam that would suggest a 5 or 6-hour online course, we can train people with no medical background to recognize glaucoma as well as local ophthalmologists. We wanted to find out whether they can serve as the backbone for a glaucoma screening program. That program involved a process of testing 15 different possible candidates and chose 2 that performed the best on the testing. Gave them self-paced training to look for glaucoma and went out into the community. In Africa, because glaucoma presents earlier, we wents to occupational health clinics and workplaces. We wanted to capture younger people. What we found is the nurses could do an excellent job. They captured the images using cameras and made their own assessment as to whether the patient had glaucoma or not and referred that patient to a central site for definitive examination and treatment as needed. Then the camera in Remidio has its own AI system. Those systems give further suggestions. Saying, well, here is what I think it is. The nurses then had the choice or the graders — they had no medical background — they had the choice if they would change their referral decision based on what AI said. We found we did very well. As well as any screening program out there for glaucoma has done. 80 to 85 percent sensitivity and specificity. We picked up 85 percent of cases. When we added the AI in, that went up to 90 percent. So very encouraging. We found that these medical, nonmedical graders with very modest training could do already a pretty good job and with the AI support from the cameras, that accuracy was improved as well. So the program has been very encouraging in terms of our ability to find glaucoma in the community in settings like Eswatini. >> I love what you said and I have a buddy that calls AI augmented intelligence. It’s assistive in helping to find glaucoma or diabetes. I’m going to shift to Dr. Charles Cleland. First, thank you for joining us. You’re doing amazing work in Tanzania. I would love to hear about the human piece. How do you keep the human piece in the loop not only for patient exchange and education but for this technology. What are some lessons that you learned from the successful program you’ve been running in Tanzania. >> Thank you for the question and thank you for inviting me today. I think the main take away from our recently completed trial in Tanzania is that the use of an AI software within this particular program basically changed the way that the screening pathway was implemented and delivered rather than specifically the AI tool itself. So basically, though the use of a camera with an AI-enabled software coupled with that, we were able to give patients their results at the point of screening. As highlighted from other panel members, this contrast with the standard way historically that diabetic eye screening has done, and images are graded at a later day by a human specialist and people have their results communicated after a number of weeks. So we found in our trial that will soon be published that the patients randomized to the AI arm of the study had an almost double probability of being connected through to the specialist eye clinic where they also had to get assessments and treatment. And this adds to the good work that professors Mathenge and Congdon do in Africa that show an AI screening program can help connect more people with potentially sight-threatening retinopathy through to the specialist clinic where they can be assessed if treated. Which of course is the ultimate thing that prevents people from losing vision from the condition. So in terms of human in the loop, I think that the use of AI task shifts away some of the human or historically human needed actions within a health system. But it certainly doesn’t replace the need for a human. A human is still needed to take an image of a good enough quality to allow the AI software to run. A human then needs to explain in a clear empathetic manner to the patient the result of the test that is being done. And counsel them appropriately on what that means and what the referral process would be. And I think in some regards, using AI within eye care and healthcare more broadly, perhaps gives — more time to interact and speak to patients rather than completely removing the human from the loop. And as professor Mathenge highlighted before, in Africa, there is a shortage of eye care specialists. So using technologies that can automate some of the clinical decision-making process has real value in that setting. >> Couldn’t agree with you more. How do we maximize the chair time so the ophthalmologist is dealing with a patient that needs an ophthalmologist and face to face time. I don’t do Facebook, I do face to Facebook. It’s important that we don’t use AI to become the next EMR where the patient spends more time seeing the back of the patient’s head than their face. We think there is a future for machine mentoring where the AI can be used to train other eye care providers or their families. So Dr. Kwesi, thank you for joining us. Super excited about the work you’re doing. Can you tell us about some of the vision you have for using AI for sickle cell retinopathy and how you envision that technology being used and what are the some of the key wins and also the challenges that you see as we go down that pathway that you’re blazing for us. Dr. Kwesi, over to you. >> Thank you, Hunter and thank you Cybersight for inviting me to talk on this. As you know, we’ve had some success with pragmatic diabetic retinopathy screening service. So we tried to tackle the elephant in the room and in west Africa, those who practiced in west Africa know that cycle cell retinopathy burden is quite huge, maybe up to 30 percent of our surgical workload is sickle cell retinopathy. The challenge with sickle cell retinopathy, it’s a disease of the peripheral retina. The standard 60-degree camera doesn’t necessarily capture most of the parts of the retina where the disease is. We have to think of an alternative way to do this. In the past, most of this has been done by skilled professionals because even using a slit lamp can be quite hard to identify some of these features so we have sent residents to the sickle cell clinic to use an ophthalmoscope to do the examination. We put grants together and wanted grants to use one of the new machines to do ultra-wide field imaging so see if we can use that as a practical step for sickle cell retinopathy screening. And it’s been not so straightforward. One of the biggest problems we have is that a lot of the pathology is quite — and fundus fluorescein angiographies are required to typify the pathology accurately. The second problem is in a low-resource setting like Ghana, it’s hard to get optos spread around the country to adequately cover the population if we’re doing a real national screening program. We are confident with some of the other machines that are not as expensive, we might be able to do something particularly where we get the clients to look in nine different positions of gaze. But this will be like a modified approach. The next issue we have is just like lack of data on sickle cell retinopathy itself. Lack of training models looking at sickle cell. Lack of good repository of images. So together as part of the pivot trial, we got the largest case series of HPSCIs that we are looking at longitudinally. We’re in year three to see if we can have a strong data set from which to be able to generate a significant model to look at sickle cell retinopathy. So these are some of the challenges and this is how we see things moving forward. >> I love — I obviously deal a lot with planes and I say the right brothers didn’t have a pilot license because they were blazing a new trail. And you’re doing a very hard thing, because you’re trying to get the data sets. And we all know that algorithms are only as good as the data sets they’re trained on. I love that you’re creating that data set. You’re doing something technically difficult with the peripheral versus the central or the macular retina. I thank you and look forward to reading some of the research you’re producing. I’m going to shift to Anusha. You have a great perspective on the entire field not just technology and implementations. The worst disease that is affecting eye care is “pilot-itis”. Everyone comes in, takes a lot of resources and do a pilot and it’s not successful or wasn’t planned for the long-term or not sustained. You have such a unique perspective, what do you see about what makes the technology last and embed itself in a long-term solution that is solving the big community eye health problem. Over to you Anusha. >> Thanks for the question. It’s a difficult question to answer. I will talk about some of the experiences that we have had. In terms of working with health systems like large health systems, both within the government as well as in the nonprofit ecosystem. I think for us, the biggest lesson is does the technology solve for a public health problem that is identified by the front-line worker. If you have developed a technology that doesn’t necessarily solve for a pain point that the health worker has, then the technology will not stick. That is one. Which requires a lot of inputs that we have to take from the front line workers and the users in the development process and only then can you develop technologies that matter. Second is even when it comes to embedding the technology in a large healthcare system, while you have a grant typically that funds for the pilot, I think the way that pilots should be designed or what we have learned from our experiences is who ..(audio blipped).. in middle income countries. There are budgets available and allocated in different programs and it is the responsibility of the technology provider along to work alongside with the ecosystem players and the professional bodies. So the professional ophthalmology counsels and national health protection schemes to make sure there are reimbursement programs that are implemented. Of course, this is a long-term perspective on what needs to happen with a developing pilot. The third thing we have learned is so — because we’re talking about algorithms and to echo Dr. — the previous panelist, the data sets that on which the algorithms are developed need to be representative of a larger population and representative of the local population that we are screening. So this involves clinical, extensive clinical evaluations done by independent groups in different geographies. Overall, I do believe that it is a continuous learning process for everyone in the ecosystem to ensure that we find the right payers, we find the right users in order to ensure the technology sticks. >> I couldn’t agree with you more. Whenever we launch technology, people are excited about the impact it has on the healthcare metric but not looking at the long-term financial sustainability. And I know you’re going to address that in the next section. I’m going to ask everyone in the audience to do a polling question. We wanted to see how you all are using AI. So I think Andy is going to send a routine poll question. Please answer it. And then we’re going to jump to the next section after we see the results. So we’re going to take a second. I do see all the questions in the chat. We’re going to have time for Q&A at the end. So please put your questions — we’re also going to put some Cybersight resources there. Pretty soon we’ll jump to the next section of the talk include is really looking at the lessons learned. So I see about half of you are not using AI in the eye screening work. I bet more than 80 percent of you are using AI for a search engine or for putting together talks. It’s very encouraging to see that about 25 percent of you all are routinely or using it in an evaluation phase. And hopefully this is the webinar that will help you to take it from evaluation to the routine. Looking at the lessons learned and thank you Andy for, that technology is something that requires a learning curve. Will is always a way that the user did not behave correctly or the technology behaved differently than we expected. Nathan, I’m going to go to you and as we talked about glaucoma is probably one of the more difficult things to screen for because we can’t get the experts to agree on the definition and get them to consistently evaluate the same. What have you seen where technology genuinely helps and where is it still falling short in the area of glaucoma. Where is the glaucoma AI or the teaching modules that we created, where is it helping? Where do we still have gaps. >> That’s a good question, Hunter. I would echo Dr. Cleland, it’s not technology in a vacuum or a lab, but how technology interacts with an existing system or model for screening. I can speak with respect to Eswatini, we made a choice. We wanted to go over younger people earlier in the disease course. There were technological benefits of that. Something like 99 percent of our participants could get an acceptable image because the mean age was 48 years old. The folks didn’t have a lot of lens opacity. The flip slide is there was a relevant prevalence, the cost effectiveness would have been lower. We had to do a lot of work to identify each person. We had one-to-one and a half percent of prevalence of disease in the people we were looking at because we were looking at young people in occupational settings. That is an issue of how the technology interacted. Other limitations with respect to glaucoma are specific to the disease. We still have good sensitivity and specificity in the mid 80s but not great. I think that is inherent to the disease. There is a lot of overlap. That becomes less if we decide to focus on people with really severe disease which I think are the folks that we need to focus on. If I’m working in a lower middle income country setting and I find someone 75 years old with a nasal step, my treatment is to go and shake their hand and say you’re never going to go blind from glaucoma. The other big piece of course that AI and machine approaches can’t help us with when we get to screening is getting the individual to come in for care. Now, we have shown that we do get more people coming through. We found that in Rwanda and Tanzania but we haven’t fully solved the problem of having people come in and certainly in terms of treatments for glaucoma, there are still challenges. We’re still cutting holes in the eye for our definitive treatment. There are definitely challenges and we can’t view AI as a complete panacea. It’s helping us and the decisions we make about how to deploy it inevitably will bring up limitations and strengths especially as we apply it to the younger population. >> I love what you said. You set Charles up nicely. Whenever we use a technology in medicine, we want to know the sensitivity and specificity and accuracy. What are the lessons you’re learning with regards to how to get it to really reach the patients and communities we want? Offline features, portable or hand held technologies what do you learn is the secret sauce for work flow and user uptake. What are some of those lessons Dr. Cleland. >> Yes. We touched on the sensitivity and specificity points. So I think this relatively early stage of deploying and using AI tools within health systems, they need to have reasonable sensitivity. So that will mean they’re not missing lots of potentially positive cases. So I think that’s the first requirement to kind of get through the door so to speak. Then, I think to give a perhaps somewhat slightly unhelpful but truthful answer, it’s going to depend a lot on the particular setting and the particular program that you are operating in. Different programs in different countries have different needs. I think that’s why it’s so important that people working locally on the ground are involved in how to actually implement and integrate these particular technologies. So I can speak with respect to our work in Tanzania which is mainly down through the Kilimanjaro Christian Center in Northern Tanzania. Some of the things that we particularly recognize as being important is one, ability to work offline. That is important. A lot of AI models are hosted on a cloud and that means that the images get taken on the local device and zipped up to the cloud which could be — the server could be anywhere in the world. If you don’t have an internet connection, that is not going to function. So offline functionality was important. And two, particularly in Tanzania, this will vary across different countries but data protection laws don’t allow for routine transfer of clinical data outside of the country which would be what would be happening if images were sent to a cloud that is hosted on a server in a different country. I think it needs to be any device and software that is going to be scaled in a low-resource setting needs to be usable by non-specialist staff. It needs to have a simple interface and be something that diabetic nurses who aren’t eye care specialists can learn fairly quickly. The cost is obviously important. Technical support from whoever the provider of the technology is and then regulatory approval is also important. Because these things are generally defined as medical devices and they need, you need to make sure you’re abiding by the kind of data governance laws in the country that you’re operating in. And then just another reflection as well in the context of diabetic retinopathy screening, the person who is taking the image and running the image through the software is also the person who is communicating the result to the patient. And that’s actually somewhat overlooked but it’s actually an important aspect of providing good quality care. You need to have someone who is motivated, who can communicate clearly, who understands the importance of what they’re doing. Because otherwise, even if you have a really sophisticated AI software with a really slick design, if you don’t have someone giving the patient the result in a really clear way, they’re never going to make it through to the clinic and be treated. I think focusing on having motivated staff who are good communicators and who are actually going to be delivering results to patients is a non-tech but important component of the whole package. >> I love what you said. I think that is one of the super powers that Dr. Mathenge has is getting reverse innovation. Not reverse engineering but hearing from the hands and the people that the workforce that are using these technologies. So looking at your experience in Rwanda, what is working well with the technology and the hands of community health and where are there gaps and things we should be listening for. >> Thank you, Hunter. Let me talk about the nurses because they’re the ones that we handed this technology to. So when we introduce the technology, the push back really came from a mix of tech anxiety I would say as well as operational friction. So some of our older nurses weren’t computer literate. So there was a genuine fear of the technology itself. In some clinics, especially the private clinic, there were concerns about how this would increase their daily workload about complications of scheduling and how it would just disrupt the established routines. There is also some confusion early on about understanding the referral guidelines. Interestingly, another barrier was by the clinic leadership. Some heads of the clinic were so protective of this new and shiny equipment they put restrictive pollicis in place that limited access by the operators. They were basically guarding this new assets. What won them over, Hunter, is No. 1, patient gratitude. So this was the biggest factor. When nurses saw how grateful the patients were, especially the screen negative ones. Because they felt they had been spared a long unnecessary journey to the eye clinic and it changed their perspective. And patient to patient testimonials popularized the intervention. Once the nurses saw this perceived value of service, it outweighed their workload concerns. Two was peer-to-peer mentoring. To overcome the tech anxiety, we didn’t try to force them to learn. We asked each clinic to identify younger, more tech comfortable nurses and we trained them and with time, they slowly trained the older ones. The ones who are less experienced with tech. So three — I think the nurses liked the program because it was straightforward. The training is just one day. Being inducted to be a screener was very quick and that made them buy into it. Ultimately, we didn’t really force adoption, we just let them build confidence through use. As they used it more and saw the direct benefit to their patients, the commitment to the technology naturally just followed. >> I think every group loves to feel included and empowered and that is one of your superpowers. I certainly want to jump to Anusha. Because you talked about this, how do we make this technology sustainable and financially viable. How do we embed this and fund this within a national health system. You’re looking how to crack that across the globe. Would love your perspectives Anusha. >> Thanks, Nathan, for the question. And following up on what we discussed in the previous answer. So what we were able to do in the specific pilot in a state government in India was to work with an independent research group that was funded by the NHS. They connected a health economic evaluation of the pilot and able to generate real-world cost effectiveness data. Which then was taken to the government. The clinical case was made by the vitreal clinical specialists and ophthalmologist societies and the health data together was able to unlock a reimbursement program as part of a national health insurance program that we have in India. So now in India, we have reimbursement for screening of diabetic retinopathy glaucoma which is a great win for not just, you know, the government, hospitals but also the private practitioners. Because now they can claim reimbursements and this also moves the burden of out of pocket expenditure from the patients and back to the government in order to pay for screening. So that’s one of the things that we’re really fortunate and very proud of in terms of being part of the ecosystem that enabled this change. So I think the larger question to ask as people working in this larger ecosystem is to understand who the payer is. Is it possible for us to unlock government mechanisms in order to make the technology as well as the entire care pathway more affordable. Clinical evidence, health economic and outcome evidence is critical in our experience to unlock these pathways. Of course, we have looked at cross subsidy models and looking at social insurance as well as having — having a certain section of the patients pay for the under- served sections. There are a lot of finance mechanisms can be unlocked depending on the region that we’re working in. This has truly been a game changer for us. >> I love what you said and with your NPH, you are looking at the system not just the technology. How does the system benefit and provide for this technology. Before we jump to the next section of this webinar, I’m excited about this next poll. Andy is going to do a poll or another self-survey of the Cybersight community. I’m really interested to see what you all are choosing as the single most important thing for implementing or choosing a screening technology. Take a minute. I was surprised by the last poll, I have my choice here and I will be interested to see what everyone picks here. Also, we are looking at the questions in the chat. So wow, I knew all of the above. We shouldn’t have included that one because it’s too easy. I knew that was going to be the vast majority: I will tell you and I think Dr. Cleland you mentioned this as well, working offline and discussing the issues of portability and hand held is important. We’re going to jump to an interesting and more controversial part of the webinar. What is not working? Where are there still gaps? There is nothing perfect in this world and we’re working with rapidly evolving technology. I would like to start with Dr. Kwesi and in your mind’s eye, what does the ideal screening solution look like for sickle cell retinopathy. And you touched on this with the imaging devices. What is currently not working and what do we need to do to fix that? Look at Dr. Kwesi, the ideal solution for sickle cell retinopathy, where do we fall short and where do we need to invest or do more innovation. >> Thank you. When you say the ideal situation, we don’t live or work in an eye deal environment. Where there are all the budgets in the world, I guess the best — I put it in every districts and there are 80 districts in Ghana. I have beautiful AI that resides on the cloud or the device. It does the initial triage. There is a way for a secondary grading. And those triaged come to hospital and are seen by a retina specialist. Unfortunately, I don’t think that’s the question you asked me. I think you mean what is a pragmatic solution. If we’re going to be pragmatic, the honest truth is with most African countries the healthcare budget cannot sustain true wide field imaging. It’s a big challenge. If anybody wants to let us know if they invented a really good camera that can do 200 degrees of field in the normal sized pupil, we would really like to talk to them about it. Because we see a big gap in the market in that area. What we are then left with is a way to deploy, we have and I know Remidio has a camera that gets to 90 to 100 degrees and there is another one that gets to 130 degrees that are not as expensive as the gold standard. The challenge we have is we will be seeing that people do not have sickle cell retinopathy when we have not examined them appropriately. That’s the chicken and the egg problem that we have. It means that we have to go for an ultra wide field camera. Do we deploy that camera to do all of the screening or target either hot spots or we target teaching hospitals to reduce the workload or we target people in specialty centers where there is some form of initial screening and then these people go in for follow up with the camera to reduce the workload in terms of follow up. It’s not an easy question and sometimes I wonder why we picked this problem. Because there is not just not resources. There is just not enough resources. Even if we had enough resources, the cameras are difficult and the image sizes are large. Trying to do it in the cloud is hard. And also we do not have a language model that works for this at the moment. So I’m sorry, I haven’t really answered the question. I have brought more problems than I’ve brought solutions. >> I disagree. You’re tackling a Mount Everest. Nathan is doing K2 with glaucoma and you’re doing Everest with sickle cell. We know about this evolving field where the retina is able to give a peak inside of the neuro vascular health, kidney disease, cardiovascular disease. I’m going to stick with you Dr. Kwesi. If you were designing a screening device from scratch. Wide field, small pupil. Offline. What would be the one feature you would love to see the new device you’re designing from scratch have and which disease would you like it to tackle? An eye disease or a systemic disease? I think we’re going to see this technology evolving very, very quickly and what we think is impossible today will be routine in 3 to 5 years. What would be your answer to a new device feature that you would love and a new disease that it got absolutely right? >> My main feature would be robustness. I want it — I don’t need the Benz or the Rolls Royce. I need the one that will start and works every day whether the current is slow or the current is high. Robustness is No. 1. The next then is it has to be an on device AI. Just because of the challenges that we have with accessing the cloud, et cetera. Those would be my top two features. >> Dr. Mathenge, I see you smiling from ear to ear, what would you like a device to have and what disease should it be tackling. >> I don’t want the Toyota, I want the Lamborghini. I want a device that once it identifies the pupil it autofocuses on the retina. We’re giving these devices to non-skilled people. And Hunter, I don’t believe in task shifting, I believe in tech shifting. The tech needs to be easy to use. I want something that all it has to do is the pupil and the device does the rest. Gets a perfectly focused image and I agree with Kwesi, the AI should be integrated. Uploading stresses people. >> Dr. Charles, I see you smirking. You’re not smiling from ear to ear but what would be the device that you would like a manufacturer to build and what problem would it solve or what disease would it tackle within the eye or systemically? >> I agree with all of — some of the points have been snapped up. One thing we found just one feature that hasn’t been specifically mentioned is a lot of people do not like having drops put in their eyes. I’m sure everyone is familiar with those patients. A device that could work well without dilation I think would be patients didn’t like that and it would make it more scalable. I think that would be good. And then in terms of one condition, I guess in relation to your oculomics, there is reasonably good evidence to support the use of AI analysis and retinal imaging for detecting cardiovascular disease which is one of the biggest killers globally. I think that would be a very exciting thing if the eye care world could make a contribution to that. >> My dad is a cardiologist and I would love to put him out of business so he can retire fully. He thought I was giving up medicine when I gave up my stethoscope. Nathan, same question, one thing you would like to see the manufacturers or device makers and what does do you want to tackle. >> I’m going to start with the disease because the disease determines functionality. I will view this differently. I think we have begun looking at the posterior poll in retinopathy glaucoma because that’s the model that really makes sense in high-income countries and investment from high-income countries, let’s be honest, that’s what’s driven the directions. Nothing against the retinopathy but cataract remains a leading cause of blindness globally. 20 years ago we did a study looking at 40 different small hospitals and followed them for three years and wanted to know what is going to help them to grow in terms of cataract surgical probability. We looked at quality and cost and patient satisfaction. There was only one variable that mattered. It was the amount of outreach screening they did. If you’re a small place with one or two ophthalmologists you can’t afford to send an ophthalmologist out into the field to screen. It needs to be accurate. If you bring in people without the disease or lose people, you lose the confidence in the community. There is a need to have good quality AI screening for cataract. It should be on a single device and robust and operate without needing access to the internet. The feature I would add in, it should have a slit beam so we can pick up nuclear cataract. A lot of folks are working on that. But it’s fair to say it’s nowhere near as advanced as we are with diabetic retinopathy. That is where AI can make a huge contribution. Allowing the smaller hospitals to do good quality screening, bringing in cataract cases without having to send their relatively few highly trained surgeons out into the field to do the work themselves. >> Absolutely. I’ve been watching Anusha’s head go up and down for each one of you. As someone designing these cameras, I’m going to add a third component to the question. What are you looking at as far as design and features, what diseases are you tackling. Going back to the economics, you’re the ones that have to solve the pain solution and getting this. Can you add the economics and what economic models you’d like to see or systems you’d like to see. Over to you Anusha. >> Thanks for the questions, Hunter. On both the systemic diseases as well as the economics. I will answer the systemic diseases piece first. We are working on AI models for pre-eclampsia and anemia. Looking at retinal images and deciphering insights for early detection of pre-eclampsia anemia among mothers. Other noncommunicable diseases in adults we’re looking at chronic kidney disease and cardiovascular risk. In the pipeline are also models for neuro degenerative disease including Alzheimer’s and dementia. And further along in the pipeline is for chronic liver disease. So the intent is to say non-invasively with a single image of the retina, are we able to gather insights into multiple systemic diseases so this allows for both the healthcare worker as well as the patient convenience as well as leveraging the power of technology in order to provide insights at the fingertips of the healthcare providers and the patients. We have been working with the gate’s foundation. On the economics and the health economic argument, I think there is very widely circulated paper on the investment case of every dollar invested in eye intervention has a 28-dollar return on the total productivity as well as quality of life. So I think one of the key takeaways — I know the audience comes from a diverse set of ecosystems. There are individual providers and they’re part of larger ecosystems. One of the things that would be nice for the eye ecosystem is focus on is health modeling. For each cases, for the technology intervention and surgical intervention and one of our learning like I mentioned, we were able to show that we have been able to reduce costs by 50%. So we brought it down from $5 to $2 in the care model specifically per patient cost. That included for the government, it cost just about $2 to not only detect for disease but also treat for disease in a large population base. That drives the argument to allocate budgets from national programs. I wanted to make that point and thanks for the questions. >> I love that you cited the investment case and Nathan and others on this call were part of this. Showing not just the financial ROI but the social ROI of eye care. So I’m going to do another poll. I see we have a ton of questions so we’re going to go to the panel. But I’m going to ask Andy to do the third survey of the audience. Especially as I speak with my colleagues in other fields of medicine and use the term oculomics and even some in academic medicine haven’t heard of it. If you see the eye as the overall window to the soul and overall vascular health. It’s something we’re all going to take forward. So far the polls have not surprised me. I’m going to see, I think I know the answer here. But I’m glad Andy didn’t do all of the above for this one. Yes, this is about, I thought it was going to be a third. And I think a lot of people it’s just like any novel technology, there is fast adopters, ones that want to see more evidence, and have others be the first to jump in the pool to test the water. I would say I appreciate the healthy skepticism. I think the hype cycle for technology, we should not all worship at the altar of technology. I certainly appreciate we have a 1/3 split. I’m going to ask everyone across the panels key takeaways. I’m going to start with Dr. Mathenge and go through the panel. What are you considering when you look at hardware and technology, especially AI in your programs, both within Rio and the screening programs. What is the one thing you’re looking at when you look at the appropriateness of hardware and AI? I will start with Dr. Mathenge and move through the rest of the audience. >> Thank you, Hunter. You know me, I always want the easiest device that gives the best image and sometimes those two are not compatible. However, besides looking at the technology itself, there are other things that are equally important and that’s what we are discovering. The necessity to close the loop. So our screening programs, we are having a challenge in tracking what happens after screening. So we know that AI increases uptake of care but to have a system that tracks where they go and what happens. I think that’s really critical. And we need to work better on that side of things rather than the hardware and software part. Sometimes we’ve chosen not to go the technology side and go the analogue side and it’s because we are thinking of things like, financial independence for programs. We don’t want programs that are totally dependent on donors: We don’t want programs that cannot integrate into the health systems with already have. And also, as Kwesi said, when you cannot guarantee national scalability, ministries of health sometimes push back on that. Because how do they decide what goes where. I diverted a bit from your question but I think it’s because we have a lot of choice in terms of hardware and in terms of software. But we shouldn’t not forget this other things that are just as important. >> I agree. You want a Lamborghini with one gas pedal but the road map and making sure you can get home and get the patients where they need to go. I know Dr. Mathenge loves everything Italian, especially the cars. Going back to your question, how do you choose quote unquote fit for purpose or good technology in your context? >> So I think I start with the fact that across Ghana are littered many broken dreams. And what I think is very important is that whatever the technology solution is, it has to integrate into the health system. Integrate into the process. It has to integrate into the way we do things already. When we try to do a wholesale change tends to be when we feel. For me, how do we use the technology to empower the big pool to do their work better, faster, safer than they’re doing it. It’s clear that there has to be some hand holding between us and the device manufacturers. So simple things like explaining to them why you have to have a UPS inside, built into the machine. Explain why you need to have an inverter so the amount of current the machine can take is higher or lower. So I come back to robustness and unfortunately robustness has to be dictated by us. Just as Nathan said, the devices are manufactured for a specific market and not for our market. I also want to touch on this oculomics thing. Recently there’s been a couple of companies that have shown how you can image the entire eye, both with OCT and with fundus images. And that would be the gold standard. AI is here, AI is going to get better. The AI we are using today is going to be the worst we have used. And once we can get the whole data set of the eye, what we can do is going to be phenomenal. The future is great and we have to wait to integrate it into our care. Charles, obviously, you have run I think one of the largest randomized clinical trials in this space. And I think Dr. Mathenge brings up the human piece. But going back to the hardware and AI, what does good look like, and what are the trials the tribulations. Especially accuse the pun about the trials. >> I think diagnostic accuracy is clearly important and having local validation in the population that you want to implement is important. One thing that we have written about and have been working on is the importance of comparative evaluations of different available devices on independent data sets. I think that is valuable for people who are thinking about deploying AI tools. And there’s a number of publications where results have been published and that is of limited value because if you want to decide which device to use and you have an anonymous set of results, that doesn’t really help you. I think that’s important. One thing I think is also just stepping back for a sec and just thinking about what we’re ultimately wanting to do here which is prevent people losing vision from different conditions, the screening models need to be coupled with good treatment models. Because screening doesn’t actually prevent someone losing vision from diabetes or glaucoma or whatever the condition is. And I think there’s work to be done on developing good quality, effective, affordable treatment services for different conditions alongside good quality screening programs and I think that’s a very important thing and shouldn’t be neglected with all the hype around AI which is exciting and has lots of important effects and positive impact. But we shouldn’t neglect that people need treatment in order to have a good vision outcome. I think that’s really important. >> Everyone here, I don’t want to do diagnose and adios. I want to make sure they get good treatment and captured within the system. I certainly appreciate what you said. Nathan, you obviously took interesting finding with Cybersight and the courses in Vietnam and brought them to Eswatini. How can people like Anusha make good technology better? What are things that you would like to see us adopt or learn from with the trials? What would you like to see technology hold for the future. >> Thanks for the question. In terms of our ideas about what we want the technology to do for us, I think our audience has led us today. They don’t want a device that is low cost or easy to use or highly accurate. They want all of the above. And I think that’s reasonable. I think it’s the idea that we’re trying to accomplish all of those things. We’ve heard it from our potential user base. One thing I would say that echoes what Dr. Kwesi is saying, my bottom line about a device I want to use is that it wasn’t designed ultimately for use in Cambridge or oxford or New York or Japan. It was actually designed for use in Malawi or Tanzania, the places where we’re working. We have heard a number of cases today about how something that works very well in one setting may not work in others. So I think that’s the bottom line. Is the devices that we want to use now and in the future in the low resource settings where we’re working should be devices designed for those settings. Designed to answer the questions and the needs of those settings. >> I love what you said Nathan and I’m going to jump to Anusha in a second. When we look at cell phone technology and around the world how many people are on cell technology. I do think medical device, remote patient monitoring, and screening devices will find it’s way to the most remote corners of the room. And so will internet as we’ve seen StarLink and other groups globalizing internet access. You have a unique perspective, you are working with different groups and different clinical trials. What are your thoughts and then I have a follow up question for you. >> So I think like what we are hearing is technology needs to be context specific. It needs to be robust and accessible and affordable. So I think technology also has — it’s of course not tend all solution like Dr. Mathenge and others have talked about, it’s only part of the larger care pathway. But at the same time I think it can also leverage and empower and improve diagnostic confidence in the front line workers and others. So it enables us to leapfrog if we can use that term, right. Because we have seen some of the lower and middle income countries, we skipped the cordless phones and went straight to smartphones. We had land lines and skipped straight, you know. There is an incredible power in dissemination when its comes to technology and we’re key on building technology that solves problems of the last mile. And that requires us to develop a holistic approach. We’re excited about the interest that the audience and the community have in using this technology to look beyond the eye. So I think that’s really like, it’s gotten us thinking the questions and the answers and the inputs from everyone has really given us a lot of fodder to work on and go back to the drawing board. >> You work with partners both within the eye space and maternal fetal and other areas. From your perspective because you have worked across continents and countries and partners and private and public, what have you seen as the accelerators and the barriers to adoption? If the people in the audience, most of the people on the Cybersight webinar are the front-line workers in eye health, what are lessons that you have seen. You have worked in all context and different countries. What works? What isn’t working. What to you wish we would know as eye care professionals from the industry? >> I think the larger conversation in the healthcare community is about cross sector collaboration. This is like an amazing opportunity for the eye health community to look beyond the eye and I would say like once the technology is available, become pivotal in being the first point of contact. It’s the eye care professionals who are the first point of contact for patients. It’s an opportune moment. Universal health coverage, all of these conversations I think pivot around the same sort of pillars like talking about infrastructure and healthcare workers and the care continuum and economics. The conversations are so similar, it’s a great opportunity for eye care to collaborate with primary care to collaborate with the individual disease specialists in order to accelerate and allow for early detection. I think it’s a great moment for the eye care community. In terms of barriers to adoption, of course there is healthy skepticism about the evidence and there should be in the case of developing medical algorithms. And it is important for us to generate the right evidence through large clinical trials, through large data sets. We are really looking forward to working with the healthcare community on this in order to develop the technology that is appropriate. >> I think you said it very wisely. I think a lot of times people, and I’m probably guilty of this, we receive a technology and we’re excited. Like a car, no one reads the operator’s manual before they start driving. They studied this technology in a clinical trial, it was done in a controlled manner and studied for all different variables. I do think the launch, just like the launch of a rocket ship is super important. If you don’t launch and set the right expectations and the right trainings and don’t do it in a controlled and monitored manner, the technology will never reach its potential. Dr. Charles, I would be interested to hear you thoughts. From your launch and you were probably one of the first in the world and in Tanzania to launch the technology, what do you see as accelerator ands barriers to adoption and what do you wish you could have gone back and told yourself to focus more or less on at the launch of that spaceship. >> I would agree with your comments there which are essentially you have a difference between a trial setting where everything is very controlled. Everybody is motivated. That’s very different to wider adoption and scaling within community settings and district hospitals or the like. I think outside of a trial setting, the context becomes much more important. You’ve established that the device can work and it can do what it’s supposed to do. But actually are people on a Thursday afternoon when the clinic is really busy and one nurse is off sick, other nurses for no extra pay, are they going to start imaging eyes and counseling patients. Are they going to do that sustainably or frustrated by it or want do it. All of the contextual factors come into play along with the other things that we mentioned already in regards to regulatory approval, data governance, what the hospital management think and offline functionality, et cetera. I think there’s a whole other area of work related to implementation of technology beyond a clinical trial setting which is quite distinct from real-world adoption. >> I agree. Our patients and our providers don’t live in clinical trials and Dr. Mathenge, I would love your perspective. You published the raider’s paper with Nathan four years ago, hard to believe. There is a big gap between the academic center of Rio where you have great people and resources and run down the hall and troubleshoot. How do you move that concentric ring and get it out to the areas it’s needed most. After what Dr. Charles did, you do the clinical trial and learn best practices and get feedback, how do you get that to the last mile of care? >> What I have learned is you have to choose where you put the technology carefully. There has to be commitment from that facility because they have to commit time, they have to commit resources whether it’s internet or whether it’s protecting the equipment. And we’ve had challenges with patient uptake. So getting patients to participate regularly can be a hurdle sometimes. Nurse fatigue, as much as they’re willing to run the technology, sometimes they — their workloads are really heavy. What I would say, Hunter is that no matter the limitations, we have not been discovered in Rwanda. We do not see this as our reason to despair or think of abandoning the program whenever challenges come. And we will persist because there is no other viable option for mass DR screening in Rwanda. And AI is what I would call our practical workforce multiplier. And therefore, we will just, you know, relentlessly refine this program and work through the implementation pains. We’ll keep empowering the nurses and informing the patients. And we want to keep the specialists where they should be which is at the treatment end and that is what this is all about. >> I think Nathan hit on that nicely. Dr. Kwesi, I want to thank you. I love what you’re doing and I can’t wait to see some of your work published in peer-review journals. For you you’ve been very specific and I appreciate that about how important it is that we don’t have good intentions and “pilot-itis” littering the countryside because it wasn’t designed with the national healthcare system or someone like you in mind. With someone like Anusha online, what is one thing you would love for her to hear from you about a barrier you are facing that she can work on with her incredible team to make your dream a reality in Ghana and across Africa? >> Thank you very much. So that’s a good question. >> I will give you two because you won the gold medal, what are some things you want the industry to hear. You heard about robustness and the technical demands of the retina. But getting this understood and advocacy and research adopted by the research or the ministry of health. Charles, you mentioned regulatory and approvals. Where is something that the incredible network and resources that Anusha has at her disposal. What is something beyond the camera or algorithm? I need more data sets or more support with advocacy and policy, I need more research and evidence. So when I meet with the minister of health. What is something beyond the camera you need to realize your vision. >> Absolutely. So Hunter, I think you’ve hit on it and I’m going to echo what Ciku said about commitment. We need help with advocacy to get commitment from the government and policymakers and commitment from payers. I’m familiar with the British National — the England National Diabetic Retinopathy Screening Service. And before the former prime minister had an eye problem and it wasn’t funded. It took an eye problem to get funded. Advocacy is one of the greatest things that we do. Anusha with our health economics background I think is a key place. I think given the committed workforces that we have, once the political commitment is there and once the government policy is there and the payers are prepared to pay for it, we will make the rest happen. We will work hard to get the data sets and work hard to do the trials that Charles and Nathan and Ciku are doing. Without the political commitment, it’s extremely hard, particularly on the national level to do these things. >> We have gone a little over time and we have over 500 people in the chat. I doe you have got some good questions in there. Anusha, what are your takeaways and how would you like to wrap up this call? We need to do more sessions and I have never seen the Q&A so active. What are you takeaways and something you want us to think about on this side of the equation as eye health providers? >> I mean that’s — it’s a hard question to answer because I wouldn’t want to state anything. Because I think it’s a really incredible community. Like a very committed community that has been driving, creates strides in terms of improving eye care outcomes over — now. In terms of what I think we can do as technology providers to support the eye care community is to have more transparent conversations across the ecosystem. I keep using the word ecosystem because I truly believe in the power of collective movements and I think industry, the eye care community, the payers, the governments, all of them together can drive real change. I think with some of the newer technologies that are coming into the fore and with what is available in terms of some of the political commitments that we’re seeing in different global health organizations, I think it’s a wonderful opportunity for all of us to coalesce together and work together. I’m really looking forward to collaborations. >> I think you nailed it on the head. The triangle between providers, patients, and industry. Who are the ones innovating. I hope you heard things to take back so you have a rugged Lamborghini in the future. I want to end as I began by saying thank you. Thank you to the panelists for making the time and the effort to be on this webinar and for what you’re pioneer in this space. I have never been more excited to do global ophthalmology. And not only having tech for tech sake but getting it to reach the communities that we care about the most. Super excited. For those who watched I have tried to wrap up some of the questions. I will follow up on some of these. I want to thank those who joined and stays extra. It was too many good things to cover. I want to thank all of the panelists and the viewers and the Cybersight community for making this possible. I’m so excited to see what next year’s webinar on AI looks like and oculomics and where we are in five years. With that I will say good day to everyone. It’s probably evening for some. I’m off to the airport. Hopefully, I get my luggage when I reach Peru. And thank you Andy for all of the support that you have given to the team and to this webinar. Bye, everyone.